The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to interpolate user-provided planning requests and generated thread names directly into shell commands, such as mfl run --stage plan "<planning-request>" and mfl thread set name '<name>'. This creates a potential surface for command injection if the input contains shell-sensitive characters that escape the quoting mechanisms.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from external sources which may contain untrusted content.
Ingestion points: The agent reads data into its context from the output of mfl thread status and mfl handoff get in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or warnings to ignore embedded instructions within the processed tool outputs.
Capability inventory: The skill possesses significant capabilities, including executing subprocesses for the mfl, paseo, and git CLI tools.
Sanitization: There is no evidence of sanitization or validation of the ingested data before it influences agent decision-making or is passed to other commands.

meow-plan